Research map / Optimization / Implicit bias and generalization
Systematic Investigation of Sparse Perturbed Sharpness-Aware Minimization Optimizer
Peng Mi, Li Shen, Tianhe Ren, Yiyi Zhou, Tianshuo Xu, Xiaoshuai Sun, Tongliang Liu, Rongrong Ji, Dacheng Tao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| sparsify the perturbation to 50% (SSAM-F) instead of full SAM perturbation | SAM, ResNet-18 · CIFAR100 | test accuracy: 81.03% -> 81.24% | Improved |
| sparsify the perturbation to 50% (SSAM-F) instead of full SAM | SAM, WideResNet28-10 · CIFAR100 | test accuracy: 84.20% -> 85.16% | Improved |
| sparsify the perturbation to 50% (SSAM-D) instead of full SAM | SAM, ResNet-18 · CIFAR100 | test accuracy: 81.03% -> 80.59% | Hurt |
| raise sparsity from 50% to 99% (SSAM-D) (ablation) | SSAM-D, ResNet-18 · CIFAR100 | test accuracy: 80.59% -> 79.61% | Hurt |
| raise sparsity from 50% to 90% (SSAM-F) (ablation) | SSAM-F, ResNet-50 · ImageNet-1k | test accuracy: 77.31% -> 76.74% | Hurt |
| use Fisher-information top-k mask instead of random mask (ablation) | SSAM-F, ResNet-50 · ImageNet-1k | test accuracy: 77.08% -> 77.31% | Improved |
| drop flattest weights instead of sharpest weights (ablation) | SSAM-D, ResNet-50 · ImageNet-1k | test accuracy: 76.68% -> 77.25% | Improved |
| drop flattest weights instead of random drop (ablation) | SSAM-D, ResNet-50 · ImageNet-1k | test accuracy: 77.08% -> 77.25% | Improved |
| use unstructured instead of structured kernel mask (ablation) | SSAM-F, ResNet-18 · CIFAR10 (50% sparsity) | test accuracy: 96.73% -> 96.81% | Improved |
| use explicit instead of implicit structured perturbation (ablation) | SSAM-F, ResNet-18 · CIFAR10 (50% sparsity) | test accuracy: 96.44% -> 96.73% | Improved |
| use implicit N:M sparse perturbation (1:2) for ViT | SSAM-F, ViT · CIFAR time per iteration | time per iteration: 4.10s -> 3.58s | Improved |
| compute perturbation on 25% of data instead of 100% (ablation) | SAM, ResNet-18 · CIFAR10 | test accuracy: 96.83% -> 96.55% | Hurt |
Closest papers by meaning
- A Single-Step, Sharpness-Aware Minimization is All You Need to Achieve Efficient and Accurate Sparse Training (NeurIPS 2024 · similarity 0.95)
- Sharpness-Aware Minimization: General Analysis and Improved Rates (ICLR 2025 · similarity 0.92)
- Sharpness-Aware Minimization Efficiently Selects Flatter Minima Late In Training (ICLR 2025 · similarity 0.91)
- Friendly Sharpness-Aware Minimization (CVPR 2024 · similarity 0.91)
- $\boldsymbol{\mu}\mathbf{P^2}$: Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling (NeurIPS 2024 · similarity 0.90)
- Momentum-SAM: Sharpness Aware Minimization without Computational Overhead (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
